Researchers at the University of California, Riverside have created a machine learning model that screened over 50 million compounds to identify approximately 130 potential bee repellents, offering a new approach to protecting pollinator populations from pesticide exposure. The interdisciplinary team, led by Anandasankar Ray, a professor of molecular, cell and systems biology, directly addressed the challenge posed by the honey bee’s complex olfactory system, comprised of more than 200 odor receptors.
“Bees rely heavily on their sense of smell to forage, but that sensitivity makes it tough to find odors that push them away instead of drawing them in,” Ray said. The team’s findings, published in eLife, demonstrate that the model accurately predicted compounds that reliably repelled bees in both laboratory and field settings.
Machine Learning Model Predicts Bee-Repelling Compounds from Olfactory Data
With more than 200 odor receptors, identifying scents that deter bees rather than attract them presented a significant hurdle, one the team directly addressed through computational methods. This approach bypassed the need for exhaustive physical testing of countless chemicals, accelerating the discovery process considerably. The research team refined the model using behavioral data gathered not only from honey bees but also from Drosophila (fruit flies), an unexpected methodological choice that ultimately improved the accuracy of predicting insect olfactory responses.
Lab tests confirmed the model’s predictive power; honey bees exhibited clear avoidance behaviors when exposed to the top-performing candidate compounds, closely mirroring the machine learning predictions. Subsequent field experiments with free-flying bees validated these findings, demonstrating that all seven compounds tested reliably repelled bees from honeycombs without causing them harm. This dual validation, in controlled laboratory settings and real-world conditions, strengthens the potential for practical application of the identified repellents.
The implications extend beyond simply protecting bees from pesticide exposure. Ray notes that reducing beehive formation in public spaces like hospitals, office buildings, and residential areas can minimize human-bee conflicts.
The repellents could be valuable in agricultural settings where pollination is undesirable, such as in the production of seedless fruit varieties. “Also, some agricultural practices want to avoid pollination altogether, particularly with seedless fruit varieties.” The team’s work represents a step toward developing bee-friendly pesticide formulations, a crucial advancement given the ongoing decline in honey bee populations and the vital role they play in global food production.
By minimizing bee contact with harmful pesticides, the researchers aim to reduce exposure risks without compromising crop protection. “By keeping bees away from harmful pesticides, we can potentially reduce their risk of exposure without compromising the protection of crops,” Ray explained. This research, Ray believes, demonstrates the power of machine learning to address complex ecological challenges. “This is a powerful demonstration of how machine learning can help solve real-world ecological problems,” he said.
He envisions this model serving as a guide for future explorations of machine learning-guided solutions in environmental protection and sustainable farming practices. “Protecting pollinators doesn’t have to come at the expense of food security,” Ray said, “and with the right tools, we can strike a balance—this model helps us get there.”
It is generally thought you need abundant data to do any kind of machine learning, but that’s not true for olfaction.
Anandasankar Ray, professor of molecular, cell and systems biology at University of California, Riverside
See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
